# The Fed's Bank Stress Test Is Wrong

|
Includes: BAC, C, CMBS, GS, JPM, MS, USB, WFC
by: Benjamin Solomon
Bonds, research analyst, Data analytics, value
Summary

I believe that the Fed is well-intentioned and is unwilling to see a repeat of the Great Recession, but its Bank Stress Test is wrong.

Why did Quantitative Easing and Geithner’s “Wall of Money” work?

I explain in very simple and clear terms how an investor should interpret loss analytics, and what type of questions to ask of the CRO.

Can the banks hold up if we are heading into arecession?

### Some Basics in Loss Analytics

Based on over a decade of my work in Commercial Mortgage Backed Securities (CMBS) and some Residential (RMBS) loss modeling, the generic loss profile is depicted in Fig. 1. Yes, I figured out how to calculate Black Swans. I have named Fig. 1, the Risk Centerfold as it depicts all useful measures of risk in relation to each other based on my first-hand experience working these metrics. Loss risk is the probability distribution of losses in dollars, given a specific reference condition, that a default has occurred and must not include \$0 losses.

As Fig. 1 shows, most losses are centered around the mode or peak of the distribution. The average or mean μ loss is always to the right of this mode as depicted by 1x. Occasionally, the losses are so large (Var, CVaR & Black Swans) they can be many times greater than the mean as this distribution is skewed to the right; i.e. the right tail is very long.

VaR or Value at Risk as a Measure of Extreme Loss (MEL – my term to facilitate clarity and bury all the complexities in loss analytics) is more frequently used in trading and rarely in securitization. VaR is usually set at the 98th percentile. The mortgage securitization industry (i.e. bonds) uses a MEL of 3x. In very simple terms this MEL value is used to determine the risk capital, a financial institution needs to have available to it in the eventuality of an adverse economic environment.

This loss distribution will shift to the right in adverse times and economies. Banks and companies do fold in good times, too, you know. If this shift to the right is significant, MELs can evolve into Black Swans. The problem with Black Swans is that they are summative and tend to domino. For the economy to be resilient to crashes, we need in place anti-domino mechanisms. Right now, my guess is as good as yours as to what those mechanisms would be. Halting an exchange is not it. Nor is more legislation. What is required is something that is market-driven.

How did the 3x MEL come about? As far as I could figure out, it has its origins in the Normal distribution. Early in the industry, losses were considered Normally distributed, and thus had symmetrical tails. It so happened that the standard deviations σ of these loss distributions were approximately equal to their means μ. Thus 1x, 2x and 3x were equivalent to number of standard deviations away from the mean μ and 3x would be approximately 3σ or 98%. The world has moved on, but the mortgage securitization industry still holds on to its old ways.

Fig. 1: The Risk Centerfold (Source: Ben Solomon 2017)

As one can see, the MEL of 3x falls far short of the MEL of VaR which is around 5x to 7x, and therefore underestimates the “real” MEL. I have termed this the Blindsided Methodology Risk, the risk we don’t measure because our statistic is short-sighted. Personally, I prefer CVaR, Conditional Value at Risk, because it is much more stable than VaR; however, it is 10x computationally more intensive than VaR.

Assuming that risk capital is set aside to cushion unexpected losses, how does this MEL cushion work? Very simply and elegantly, risk capital shifts the loss distribution to the left. That is why, Timothy Geithner, the Secretary of the Treasury during the Great Recession, used a “Wall of Money” to substantially reduce the risk to the economy. See Stress Test. Yes, Quantitative Easing was the right thing to do at that time. It was a very innovative and powerful mechanism to transfer cash to the banks and shift the banks’ loss risk distributions to the left.

Note, however, risk capital only converts “unabsorbable” losses to absorbable losses and does not change the risk profile of the bank. Risk capital is a balance sheet play, while the risk profile (shape of the loss distribution) is an income play. Therefore, to infer a bank’s risk profile I would suggest using Income Before Fees per unit of Assets (IBFA).

### What About the Fed’s Bank Stress Test?

The point of any stress test is to ensure that the company undergoing the test survives a severely adverse environment. Therefore, MEL cannot be some average “bad” number but should be the worst possible outcome, say 98% of the time. The June 21, 2018, Stress Test shows that bank capital would fall 4.4% from 12.9% to 7.9%. What 4.4%? Wait a minute, that was a result I was getting when I was new to CMBS VaR & CVaR loss analytics in 2003. Another bank (I’m not allowed to reveal their name) I knew, with a young risk team had 3% - they had not taken into account the long tail. I had to check.

Using the latest Fed’s Bank Stress Test results (their model is very thorough but don’t mistake thoroughness for correctness) I compiled what the Feds used for their Severely Adverse case versus worst scenarios in recent history. I ignored the really big test of 1929 as the Fed had ignored that, too. I didn’t bother with their Adverse case. Just using the statistics I’m familiar with, the tables below show what actually happened in recent history versus what the Fed used.

 Table 1: GDP Growth – is grossly underestimated Actual Stress GDP From 1Q78 16.40% From 3Q17 3% To 1Q80 -8% To 2Q18 -9% Change -24.40% Change -12% Table 2: S&P/Case-Shiller U.S. National Home Price Index (data limited) – probably underestimated as California had much higher loss rates of -40.9% between 3Q 2006 and 2Q 2012. Actual Stress Test From Jul-01 184.6 From 4Q2017 To Feb-12 134.0 To 3Q2019 Change -27.41% Change -30% Table 3: Unemployment – may be OK Actual Stress Test From May-79 5% From 4Q2017 4% To Apr-83 11% To 3Q2019 10% Change -6% Change -6%

The Fed’s Bank Stress Test does underestimate worst-case outcomes. Here is why:

1. Primary Flaw: Statistical models tend to revert to the mean, i.e. these provide averages not outliers as is required of long tails.
2. Not Handling Catastrophes: Catastrophes occur when a few unrelated events come together. (Look up Malcolm Gladwell’s Outliers). Therefore, using statistical input models to determine related inputs severely reduces the possibility of worst-case events coming together.
3. What Risk Methodology?: It is not clear what risk methodology the Fed is using. Though not documented, my initial inference was that the Fed may have used a 1x, 2x, 3x approach, but I could not even find that. I infer that the Bank Stress Test originated in 2009, in the rush to calm the public, just after the Wall Street Crash of 2008, and during the Great Recession. The Fed cobbled together some econometric models and showed an output that said everything was okay. That cobble is now the standard.
4. Missing Distribution: There are no Monte Carlo outputs, and the Fed report does not discuss what percentile the 4.4% belongs to. Imagine you have \$13.1 trillion in assets and you don’t even discuss what percentile 4.4% belongs to?

### The Correct Approach to Stress Testing

The correct approach to conducting a stress test is to frame the problem in terms of when the bank will default and then work backwards to determine what it takes to keep the bank solvent. The Feds avoid both these questions.

The next step is to identify the range of input values for each input variable. Determine their probability distributions or at least state your distribution assumptions. Determine possible correlations between inputs but be careful how this is factored in.

Recall that the loss distribution is based on a specific reference condition, the default, and that loss must be greater than \$0. Similarly, in bank stress test, the reference condition is, bank default is eminent when bank capital declines below some specific value; say 5%, and when the economy is good. The lower this number, the safer the bank! Why do I say, “when the economy is good”? Because stress testing is about whether the bank (or for that matter any organization) will hold up, going from a great economy to an adverse one.

Also note that capital structure is primarily about the bank’s cost of maintaining that capital. The free market does not care what you paid for that structure, just that you pay.

Feed these input data into a bank Monte Carlo model. Plot the distribution of the statistic of interest –bank capital and only when that capital is below that reference value. Determine the additional capital, below this reference condition, required to cover the 98th percentile worst case.

Therefore, the Fed’s Bank Stress Test should report three parameters,

1. CapSolv: Minimum capital required to keep a bank solvent in good times. CapSolv is the reference default condition for bank stress testing. It shows how well a bank’s leverage is structured, after all they are borrowing from the public.
2. CapAbv: Capital required above/greater than CapSolv, required to cushion losses in good times, as banks do fold in good times, too. Reported at the 98th percentile. CapAbv shows how well they can weather a rash of bad events (e.g. major borrower like Enron, WorldCom, GM . . . going bust, etc.) in good times. CapAbv should be similar across banks as it is very much an indicator of factors in the external economy. I won’t be surprised if there are regional or state variations in this statistic.
3. CapRisk: Risk capital required to cushion losses in adverse times, reported at the 98th percentile. CapRisk shows how risky the bank’s policies are. Given that CapSolv < CapAbv < CapRisk, a bank’s capital structure must be at least that of CapRisk.

Very good banks have very low CapSolv, and the smaller the difference between CapRisk and CapAbv for a given CapSolv, the better the bank’s risk profile.

I believe that the Fed is well-intentioned and unwilling to see a repeat of the Great Recession, but from a risk perspective, I have no idea what the Feds are reporting with the Bank Stress Test. It is irrelevant as it is wrong.

### What Should Investors Ask at Investor Meetings?

You don’t attend an investors’ meeting and ask the Chief Risk Officer if his trucks are reliably maintained to deliver goods? Or if his warehouse can securely hold the commodities, he is long on? Do you? No kidding, I actually heard that on a conference call. Those are questions for the COO.

The following are sample risk-focused questions to ask.

1. How did you arrive at your MEL figures?
2. How stable are your MEL figures?
3. How have you physically implemented your MEL cushion?
4. By how much would these MEL estimates increase if the economy deteriorated?
5. Could these losses evolve into Black Swans?
6. What cash flow management (not cash flow reporting or forecasting) do you have in place and how frequently does your management team review this? (A US Bank study found that 83% of corporate failures show up as cash flow problems i.e. cash flow is on the frontline of detecting problems.)
7. How does adding the new product alter your MEL and what precautions have you taken to mitigate possible losses?
8. What is the lowest capital the bank or financial services company can operate with, without defaulting on its obligations?
9. What is your Net Income Before Fees per dollar of Assets? And how does it compare with other banks in your peer group?

### What Should Investors Do?

Go back to the basics. Look at the banks’ Net Interest Margin (NIM) and Income Before Fees per Assets (IBFA), after all that is the primary purpose of banks. They are the intermediaries of funds, borrow from one group and lend to another. Note getting to IBFA is not easy as it requires a lot of digging. Compare with their peers. If it is higher than its peers, find out why. The answer maybe that they have riskier assets and borrowers are likely to readjust, i.e. it is not sustainable. If lower, again find out why. Are they growing by discounting?

Obviously by now, you know that I am an independent thinker and neither a conformist nor a contrarian. With that in mind here is a proposal on how to analyze bank stocks, without the benefit of knowing their fee income.

Bear in mind this is a quick and dirty analysis only. Using Friday’s (12/28/2018) closing prices, I put together a quick analysis of the 7 biggest US banks. See Table 4.

For banks, gross profit which is interest income is essentially revenue. Thus, revenue per dollar of asset is a measure of the riskiness of the bank assets assuming risk-reward is linear.

 Table 4: Bank Data Chase BofA Wells Fargo Citigroup Goldman Sachs Morgan Stanley US Bancorp Revenue (\$B) 93.689 83.956 85.989 64.277 32.073 37.945 20.038 Gross Profit (\$B) 93.689 83.956 85.989 64.277 28.636 34.061 20.038 Share price (\$) 96.83 24.39 45.78 51.83 163.03 39.37 45.21 Market Cap (\$B) 321.954 239.368 215.497 126.576 60.642 67.722 73.063 MktCap(\$B)/Asset(\$B) 0.1271 0.1049 0.1104 0.0687 0.0661 0.0795 0.1581 Assets(\$B) 2,534 2,281 1,952 1,842 917 852 462 Gross Profit / Assets 3.70% 3.68% 4.41% 3.49% 3.12% 4.00% 4.34%

Source: Yahoo Finance 2018, bank data based on latest annual filings.

Okay, this is difficult to make sense of, right? Draw graphs, then the picture becomes clearer.

Fig. 2: Revenue versus Assets (Source: Ben Solomon 2018)

Fig. 2 depicts revenue produced by bank assets. It is clear that all banks behave similarly. For every dollar of assets, they generate \$0.0376 of revenue. Therefore, revenue growth is achieved by asset growth, not by product innovation. That is, there are no scale efficiencies to be gained at this end of the banking segment.

Fig. 3: Revenue per Asset versus Total Assets (Source: Ben Solomon 2018)

From Fig. 3 Wells Fargo (NYSE:WFC), US Bancorp (NYSE:USB) and Morgan Stanley (NYSE:MS) were generating more revenue per asset than their peers, which means their assets were riskier than the baseline risk banks of Goldman Sachs (NYSE:GS), Citigroup (NYSE:C), BofA (NYSE:BAC) and Chase (NYSE:JPM).

Fig. 4: Market Cap per Asset versus Revenue per Asset (Source: Ben Solomon 2018)

However, Fig. 4 shows that as of Friday close, the market was rewarding BofA, Chase and US Bancorp a share price premium over the other 4 banks (Goldman Sachs, Citigroup, Morgan Stanley and Wells Fargo). As such these 4 banks - Goldman Sachs, Citigroup, Morgan Stanley and Wells Fargo - are likely to move in tandem. US Bancorp is an island on its own, high risk and high rewards. With Morgan Stanley the market has factored in its risk, and it performs like the others.

This is about BofA versus Chase. Both share prices should lower, otherwise risk arbitrage is present. It is more likely that Chase's share price will decline to be equivalent to BoFA’s and match the risk profiles of the other banks and not BofA's share price increasing to match the equivalent of Chase.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.